AI-Assisted Decision-Making: An Experimental Study of Bargaining in the Second-Hand Market

How to cite

Deng, Weiguang. 2026. "AI-Assisted Decision-Making: An Experimental Study of Bargaining in the Second-Hand Market." AEA RCT Registry. AEARCTR-0017607. https://doi.org/10.1257/rct.17607-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Weiguang Deng*

* Primary investigator

Registry ID
AEARCTR-0017607
DOI
https://doi.org/10.1257/rct.17607-1.0
First registered
Last updated
Status
On going
Trial dates
2025-12-30 – 2026-05-31
Countries
Private
JEL codes
C78 (Bargaining Theory; Matching Theory), C93 (Field Experiments), D03 (Behavioral Economics: Underlying Principles), D82 (Asymmetric and Private Information), D83 (Search; Learning; Information and Knowledge; Communication; Belief; Unawareness), L81 (Retail and Wholesale Trade; e-Commerce), L86 (Information Goods and Services, Internet Content)
Registry record
http://www.socialscienceregistry.org/trials/17607

Abstract

We study whether generative AI improves bargaining outcomes in real second-hand market transactions. We conduct a randomized field experiment on Xianyu in which participants, acting as buyers, contact real sellers of listed items. A total of 2,000 buyer–seller contacts are randomly assigned to one of four experimental conditions. Primary outcomes include successful bargaining, final negotiated price and the bargaining discount relative to the posted price. This study integrates AI-assisted decision-making with the classic economics and behavioral science issue of bargaining and negotiation, provides randomized controlled evidence in real-world platform settings, helps clarify the relative importance of information supplementation and strategy substitution, and explains the gain and loss mechanisms of AI-generated content in interactive games from the perspectives of algorithm aversion and human-AI collaboration.

Intervention

A randomized controlled field experiment is conducted on the Xianyu second-hand marketplace. The experiment assigns real buyer–seller contacts to different conditions that vary the role of generative AI in the bargaining process, with the interventions designed to manipulate the degree to which AI supports decision-making and communication (e.g., informational support versus strategy support, and the extent of human involvement). Participants act as buyers, contact real sellers, and negotiate under standardized rules with a fixed maximum number of bargaining rounds and pre-specified stopping rules. The study covers multiple product keyword categories that differ in price transparency to examine how AI-assisted bargaining performs across market environments.

Experimental design

A randomized controlled field experiment is conducted on the Xianyu second-hand marketplace. The unit of randomization is an individual buyer–seller contact (a specific listing contacted by a participant). Contacts are assigned to different conditions that vary the role of generative AI in the bargaining process (e.g., whether AI provides informational support and/or assists with drafting messages, and the degree of human involvement). Bargaining follows a standardized protocol with a fixed maximum number of bargaining rounds and pre-specified stopping rules. The study includes multiple product keyword categories that differ in price transparency to examine whether treatment effects vary across market environments.

Design details

Randomization method
Computer-based randomization conducted by the research team using a random number generator (e.g., a scripted random assignment tool). Randomization is implemented within each participant × keyword block: for each keyword, the participant selects four eligible listings randomly, and the four experimental conditions (T1–T4) are assigned to these four listing-contacts without replacement (each condition used exactly once within the block).
Randomization unit
The unit of randomization is an individual buyer–seller contact (i.e., a specific listing contacted by a participant). Randomization is blocked at the participant × keyword level (within each participant and keyword, treatments are assigned across the four contacted listings).
Clusters
100 participants (buyers).
Observations
2,000 buyer–seller contacts / contacted listings (individual listing-contacts).
Arms
500 listing-contacts in each arm (total 2,000): T1 (Human-only): 500 listing-contacts T2 (AI information only): 500 listing-contacts T3 (AI information + AI messages, no edits): 500 listing-contacts T4 (AI information + AI messages + human edits): 500 listing-contacts
Primary outcomes
1. Effective bargaining (indicator) — whether the interaction results in a meaningful price concession or a negotiated price. 2. Bargaining discount (share) — the percentage discount relative to the posted price.

Cited by

No citation data available for this registration yet.

Keywords